Machine learning in trading: theory, models, practice and algo-trading - page 3465

[[Deleted]]

Aleksey Vyazmikin #:

Is there such a formula?

Well, there should be.

There's a lot of them, here's an interesting article by the way.

because I don't know what it's for, I've never done anything with it.

[[Deleted]]

Aleksey Vyazmikin #:

Anything.

Radical change of the interface is an idea for those who like to suffer :))))))

I have similar problems, but there are times when a function needs to be slightly changed - and making a separate function in the library for this purpose or overloading with settings is not very good.... so I always look for compromises.

Yes, there are also a lot of one-type functions with small changes for each algo.

At some point I started to forget what is where and you will never find anything at all. If you add version control with small changes to it, you will never find anything at all :)

and it's a pity to throw it away, as they say

[mytarmailS]

I failed to train the model on a set of signs, I did not take into account one nuance of my combinatorics, in fact, the signs turned out to be tens of millions, while for me to digest such a number of signs at once does not work. I could still do 2 million.

But I tested my usual patterns on new data, I made something like the concept of a Random Forest, I have 200 patterns, the trigger on a deal is when more than 100 patterns worked at once.

The trades on the new data turned out to be only 5 pieces for some reason, all in plus, stop/stake fixed, takeout 25% more than stop.

[mytarmailS]

I don't know if it makes sense to move forward with it...

although 5 one-way trades on one instrument, if you make more purchases, not only sales and on all pairs, you might get something.

[[Deleted]]

mytarmailS only sales and on all pairs, you might get something.

There is, spinning sets of patterns, going through combinations, it's a terribly tedious approach really, that's why I was afraid to even start :)

Or come up with a quick scheme. It's all about speed in the end.

[mytarmailS]

Maxim Dmitrievsky #:
There is, spinning sets of patterns, going through combinations, it's a terribly tedious approach really, that's why I was afraid to even start :)

Or come up with a quick scheme. It's all about speed in the end.

Started a search for patterns on buy, left the computer overnight, in the morning will be the result and will make a backtest of more purchases

[Maxim Kuznetsov]

mytarmailS #:

Yeah, well, that's understandable, but why?

Isn't there an off-the-shelf one?

If you like to pick your guts and write 500 lines of code to sum up A+B, then keep sitting on µl, why write in a high-level language?

And write 100 times already written from scratch instead of calling from one code and doing something deeper.

almost guessed :-)

There is NO identical, "standard" zigzag. One that is guaranteed to mark data equally and correctly in MQL, Python, or anywhere else. And there are a bunch of requirements to use in analysis....

[mytarmailS]

https://youtu.be/c-yf4nLgq2Q

[mytarmailS]

Maxim Kuznetsov #:

almost there :-)

There is NO identical, "standard" zigzag. One that is guaranteed to mark data equally and correctly in MQL, Python, or anywhere else. And there are a bunch of requirements to use in analysis....

You're writing

[Aleksey Vyazmikin]

Transferred the ZZ code to MQL5.

In general, the trick of this ZZ is to ignore outliers.

Maxim Kuznetsov #:
There is NO identical, "standard" zigzag. It is guaranteed to mark data equally and correctly in MQL, Python, or anywhere else. And there are also a bunch of requirements to be used in analysis...

Exactly - the algorithm must be reproduced from different places!